Rebound effects in GB road transport Steve Sorrell and Lee Stapleton November 2015
Topics Project outline Methodological challenges Estimating direct rebound effects in GB personal automotive transport Next steps
Mechanisms Reinforcing socio-technical system of automotive transport Embodied energy Changing prices, investment, employment, trade etc. Evolving infrastructure, land use, industries, norms, practices etc. that entrench car dependence and displace alternatives Lower fuel bills Savings Holiday in Spain More energy Less energy Lower running costs Drive further and more often in emptier cars Purchase larger and more powerful cars More energy
Implications Technical improvements reduce the energy cost of road transport, thereby encouraging increased transport activity and hence energy use More people travel further and more often in larger, faster, more powerful and emptier cars More goods are moved over greater distances in larger and more powerful trucks, encouraging more consumption of more and different types of goods The system of automobility of people and goods is reinforced But establishing causality is difficult when data is limited, feedbacks abound and everything is endogenous
Questions What proportion of the potential energy and carbon savings from improved fuel efficiency in GB road transport have been taken back by various types of rebound effect over the last 45 years? What mechanisms have been responsible for these effects, what factors influence their operation and outcomes and how have these changed over time? How robust are our quantitative estimates of rebound effects and how can confidence in these estimates be improved? How may these rebound effects be expected to evolve in the future? What are the implications for UK energy and climate policy
Methodological issues when estimating direct rebound effects
Direct rebound for personal automotive transport Single energy service and single energy carrier Accounts for large proportion of total oil consumption and is relatively price-sensitive Good-quality, aggregate time-series data available but measurement errors, petrol/diesel, company cars etc. Disaggregate data sometimes available from surveys Growing number of studies with diverse methodologies Most estimate direct rebounds in the range 10-30% Less energy Lower running costs Drive further and more often in emptier cars More energy
Rebound as an efficiency elasticity S = km travelled; E = Fuel use - GJ; = Fuel efficiency km/gj E S / Elasticity of fuel use with respect to energy efficiency : ( E) ( S) 1 Rebound = (S) Rebound may be defined as the elasticity of distance travelled with respect to fuel efficiency. But many datasets give only limited variation in (or don t measure) fuel efficiency.
Rebound as a price elasticity p S = fuel cost per km; p E = Fuel price; ps p E / E Elasticity of fuel use with respect to energy efficiency : s = Fuel efficiency ( pe / ) ( E) ( S) ps 1 Rebound = ps (S) Rebound may be defined as the negative of the elasticity of distance travelled with respect to fuel cost per km. Gives more variation in the explanatory variable.
Endogeneity and asymmetry Fuel efficiency may be endogenous: influenced by fuel prices influenced by expected travel demand Usually lack both a suitable instrumental variable for fuel efficiency and sufficient data for simultaneous equation model Fuel prices more likely to be exogenous, but response may be asymmetric and may differ from response to changes in fuel efficiency
Implications Efficiency elasticity of distance travelled is the natural measure, but data may not be available, there may be limited variation in this data and efficiency may be endogenous Only a few studies have obtained significant estimates for efficiency elasticities So which to believe?? Fuel cost or fuel price elasticities of distance travelled typically give more precise parameter estimates, but these are only equivalent to the efficiency elasticity under certain assumptions Lots of studies have obtained significant estimates for price elasticities
Our response Estimate three different elasticities and compare the results ( S), ( S), ( S) p E p S
Estimating direct rebound effects in GB personal automotive transport
Summary of method Econometric analysis of aggregate time series data on GB car use, fuel use, household income, congestion etc. over the period 1970-2011 Multiple specifications, different normalisations extensive robustness tests, weighted results Direct rebound effect estimated from elasticity of distance travelled with respect to: 1. Fuel efficiency 2. Fuel prices 3. Fuel cost per kilometre
Model groups Different specifications of the fuel cost of driving: Type A fuel price and efficiency: Type B fuel cost per kilometre: ( S), p ( S) E (S) ps Different specifications of distance travelled: Vehicle kilometres Passenger kilometres Different normalisations: Per capita Per adult Per licensed driver
Model groups Group Explained variable Normalisation of explained variable Specification of the fuel cost of driving 1 VKM Per capita Type A 2 VKM Per adult Type A 3 VKM Per driver Type A 4 VKM Per capita Type B 5 VKM Per adult Type B 6 VKM Per driver Type B 7 PKM Per capita Type A 8 PKM Per adult Type A 9 PKM Per driver Type A 10 PKM Per capita Type B 11 PKM Per adult Type B 12 PKM Per driver Type B
Base models Static model: ln S C t t 0 1 lnyt 2 ln pe 3 ln t 4 X t 5 t u t Dynamic model: Add a one period lag of the explained variable (S t-1 ) Variables: Vehicle or passenger km by cars (S) Mean equivalised household income (Y) Mean fuel price in ( /GJ) (p E ) Fleet average fuel efficiency (km/gj) ( ) Oil shock binary dummy (74 and 79) (X) Congestion proxy (e.g. road length/capita) (C)
Model variants Quadratic income variants: Proxy for factors contributing to peak car Asymmetric variants: Proxy for induced technical change, irreversible investments, habits etc. Reduced variants: Remove insignificant variables Co-integrated variants: Based on (low power) unit root tests; static specification, CCR method Selection between models based on robustness tests
Diagnostic tests Coefficients: do they behave? [3 tests] Residuals: do they behave? [3 tests] Stability: are predictions stable? [2 tests] Parsimony: is their a sound balance between model fit and model complexity? [3 tests] Functional form: is the model structure appropriate? [2 tests] Results used to create a composite robustness indicator to guide model selection
A lot of models.. 108 models 54 Type A (fuel price and fuel efficiency elasticities) 48 static models 48 dynamic models 54 Type B (fuel cost elasticity) 12 co-integrating models 1. Simple mean of statistically significant estimates 2. Invariance weighted mean of all estimates
Vehicle kilometres
Passenger kilometres
Occupancy
Fuel cost of driving
Equivalised income and congestion proxies
Main Results - 1 Using the fuel efficiency elasticity of distance travelled, we find little evidence of a long-run direct rebound effect over the last 40 years Using the fuel price and fuel cost elasticities of distance travelled, we find good evidence of a direct rebound effect in the range 9-36% Half of Type A models produced significant estimates of fuel price elasticity and three quarters of Type B models produced significant estimates of fuel cost elasticity Simple mean of statistically significant fuel price and fuel cost elasticities suggests a direct rebound of ~19% Invariance weighted mean of all fuel price and fuel cost elasticities suggests a direct rebound of ~16%
Means of rebound estimates Rebound measure Simple mean of significant estimates Invariance weighted mean of all estimates Average Fuel prices VKM 17.2% 15.2% 16.2% Fuel prices PKM 17.4% 15.2% 16.3% Fuel costs VKM 18.7% 15.2% 17.0% Fuel costs PKM 20.8% 17.7% 19.3% Average 18.5% 15.8% 17.2%
Main Results - 2 Estimates are slightly lower when distance travelled is normalised to the number of drivers rather than the number of adults or people Estimates are slightly higher when rebound is estimated with respect to fuel cost per kilometre, rather than fuel prices (expected) Estimates are slightly higher when using simple mean of statistically significant results, rather than invariance weighted mean of all results But significant overlaps in the rebound estimates for each specification and measure Little evidence for asymmetric responses to price / efficiency changes Little evidence in favour of dynamic over static models Little evidence that consumers respond in the same way to lower (higher) fuel efficiency as to higher (lower) fuel prices (Wald tests)
Implications Efficiency improvements have lowered the cost of car travel in GB over the last 40 years This has encouraged increased driving, which in turn has eroded around one fifth of the potential fuel savings Results are consistent with those from other studies However, the results also raise questions about the use of price elasticities to estimate rebound effects Multiple caveats and considerable scope for further work but really need disaggregate data sources Also, direct rebounds are only one part of the picture
Next steps
1. Peak Rebound Hypothesis: rebound effects have fallen over time and with increasing incomes implying greater potential for future energy savings Method: Similar to earlier study Identifying changes in elasticities over time and/or with increasing income Covariates informed by literature on peak car (e.g. urbanisation, internet use demographics etc.)
2. Service quality rebound More driving in bigger and more powerful cars Method: Construction and econometric analysis of aggregate time series of car use etc. including power and weight of vehicle stock. Rebound estimated from elasticities of weighted service demand with respect to technical efficiency or cost Weights could be power, weight or engine capacity (e.g. kw km)
Improvements in technical efficiency of new cars (litres/km)/kw Source: Schipper (2010)
Offset by increasing power of new cars (kw) Source: Schipper (2010)
3. Indirect rebound More driving and more money to spend on other stuff Method: Econometric analysis of household expenditures linked to multiregional input-output modelling. Estimate system of demand equations and simulate energy/emission consequences of respending cost savings Four previous studies with Mona Chitnis 80% 70% 60% 50% Extensions: incorporate energy services directly within the demand model? 40% 30% 20% 10% 0% Gas Electricity Vehicle fuels 1, 2 and 3 in combination (equal %) Substitution effect Income effect
4. Freight rebound More and larger trucks carrying more stuff over longer distances Method: Econometric analysis of aggregate time series data on UK road freight, manufacturing output etc. Rebound estimated for different measures of efficiency and from different elasticities Builds upon earlier decomposition analysis
5. Economy wide rebound More driving, more money, lower prices, changed investments, changed trade patterns Led by Karen Turner and colleagues, University of Strathclyde CGE modelling of UK economy linked to multiregional input-output model. Allows economywide adjustments in expenditures, prices investment, trade patterns etc. to be explored Covers both road freight and passenger transport
6. Rebound and automobility